Hi Product Hunt 👋 I’m Bakker, founder of MMTNEXUS. We built MMT Token Optimizer around one question: **Can we reduce LLM prompt and context usage without removing the facts, instructions, and outcomes the model actually needs?** Instead of treating a shorter prompt as success, the optimizer follows a guarded flow: **Analyze → Reduce → Validate → Run → Measure** Our published lab cases across 7 provider routes currently show measured input-token reductions ranging from **6.54% to 22.17%**, including: • Anthropic: 22.17% • Google Gemini: 18.21% • OpenAI: 17.91% • OpenRouter: 17.91% • DeepSeek: 17.14% • Mistral: 17.04% • xAI: 6.54% These are measured test cases, not guaranteed savings for every workload. The area I’m most interested in is **AI agents, RAG, long conversation history, and high-volume API workflows**, where context can grow quickly. For perspective, an agent workload processing 1M input tokens/day would avoid about **5.37M input tokens/month** at a 17.91% reduction. We’ve opened the product as a limited free public beta so developers can test it with their own workloads. I’d especially appreciate feedback on: • workloads where optimization works well • cases where the system should refuse to reduce context • agent/RAG edge cases • what evidence you would want before trusting optimization in production Thanks for testing it — and please challenge the methodology. That feedback is more valuable to us than simply getting another signup.

MMT Token Optimizer

Reduce AI token usage without losing the outcome

Reduce AI token usage without losing the outcome